Read on Twitter
Had a chance to take a deep dive into a recent paper on #frailty in #TAVR by @glen_martin1 @mmamas1973 et al so I figured I'd share some thoughts tweetorial-style 
🅿️ Patients: 2,624 TAVRs from the population-based UK registry... lucky them - 95 prospectively collected variables on all TAVRs in the land
ℹ️ Independent var: F-R-A-I-L-T-Y as measured by 3 scales - CSHA Clinical Frailty Scale, Katz activities of daily living scale, poor mobility per MD's impression... I personally like all of these but *caveat* they reflect disability > frailty
©️ Confounders: The expected/appropriate covariates were adjusted for... lowish risk of residual confounding or overfitting (for a change)
🅾️ Outcomes: Let's start with 30d mortality - all 3 scales predictive in univar but only Katz and poor mobility remain in multivar... AUC analysis shows incremental discriminatory value of frailty when added to traditional models (FRANCE-2, Observant, or ACC/TVT)
Sidebar: what is the AUC anyway?
A. Area under plot of sensitivity vs. 1 – specificity
B. Given a pair of individuals, the AUC is the probability that the predicted risk will be higher in the one who develops an adverse event vs. the one who does not
C. C-statistic
A. Area under plot of sensitivity vs. 1 – specificity
B. Given a pair of individuals, the AUC is the probability that the predicted risk will be higher in the one who develops an adverse event vs. the one who does not
C. C-statistic
🅾️ Outcomes: 1-2y mortality - again all 3 scales look good in univar but Katz and poor mobility appear to have higher HRs in multivar (stronger effect after adjusting for confounders)
Devil is in the details / supplemental appendix ... my back-of-the-envelope number crunching suggests that CSHA cutoff >=6 (moderate frailty) yields higher OR of 2.53 for 30d mortality... would have liked to see a sensitivity analysis w/ scales as ordinal instead of dichotomous
Final thoughts: nice study with exemplary manuscript transparence... Katz disability and poor mobility easily assessed and highly predictive patient-centered measures but they can be subjective... ordinal > dichotomous... ideally add-on physical and cognitive performance measures
Correct answer was... ALL OF THE ABOVE ☑️
• • •
Missing some Tweet in this thread? You can try to
force a refresh
